Predicting Earnings Surprises

126 Pages Posted: 17 Jul 2019 Last revised: 21 Jul 2019

Date Written: July 27, 2017

Abstract

Nonlinear classification models, using just pricing and earnings information, can with great success predict future earnings surprises. Surprises that deviate 15% or more can be predicted with 71% accuracy. These predictions can be used to earn abnormal profits. A machine learning regression model that incorporates pricing, analyst forecasts, and earnings information provides a better estimate of future earnings per share than that of consensus analyst forecasts and mechanical time-series models alone; with errors as much as 46% lower than analysts over certain periods. The machine learning model uses signal-processed and earnings-related variables. Some of these variables have been noted in past research to be related to analyst bias. The machine learning model in effect corrects for analyst mistakes and biases by incorporating these variables into a nonlinear prediction model that lowers the overall forecast error for all periods concerned. Machine learning models aid the identification of analyst biases by identifying variables that are important in predicting earnings over and above analysts' consensus forecast.

Keywords: Machine Learning, Earnings Surprise, Event-driven, Trading Strategy, Prediction

JEL Classification: C32, C38, C45, G14

Suggested Citation

Snow, Derek, Predicting Earnings Surprises (July 27, 2017). Available at SSRN: https://ssrn.com/abstract=3420722 or http://dx.doi.org/10.2139/ssrn.3420722

Derek Snow (Contact Author)

FirmAI, UoA, NYU FRE ( email )

NYC, Cambridge, Auckland

HOME PAGE: http://www.firmai.org

Register to save articles to
your library

Register

Paper statistics

Downloads
365
Abstract Views
1,357
rank
81,240
PlumX Metrics